Too much AI Today: Talking loud and saying nothing

"I'm perfectly willing to concede that I've been too nice in thinking that if you had a thousand parameters maybe you'll get somewhere. maybe you won't. but my concern is different. if the system doesn't distinguish what's the actual World from non-actual world it's not telling us any thing. just as in the case [that] the souped-up periodic table the [deep learning?] systems and others like them can find superficial regularities in astronomical amounts of data and produce something that looks more or less like what their data was but it can do exactly the same thing even better often with data that violates all the principles of language and cognition. so they're just telling us nothing." - Noam Chomsky, 2022

I haven't been too much about Chomsky to date. And now he is in the later innings. He had something to say about AI this month, which I thought valid to share. Important because this has so much been the year of Generative AI news. So much of AI's success in recent years has been about Language Processing. But we shouldn't now get to carried away with where these successes may take us. I thought Chomsky distilled this in his commentary broadcast to attendees at the Web Summit in Lisbon.

Chomsky seems fair minded. He appreciates the tool that allows him to receive questions of interviewers, despite his hardness of hearing. A good utilitarian use that. But to claim more, no.

But an AI model is just that. As Facebook found just last week. It rolled out Galactica, a large language model that learns to write text "by studying millions of examples and understanding the statistical relationships between words." It doesn't actually  understand meaning - it's what some call a "stochastic parrot." 

A nice VentureBeat piece I read recently by Ben Dickson note the upside of generative AI, but also the downside. These "AIs" are capable of unpredictable mistakes. Much touted DALL-E model does require human agency, it is said.

Have to admit a lot of the flotsam about is really galling to me. I think I hold some right to my annoyance having covered AI (off and on, admittedly) since 1985. 

Final Aside: DALL-E is an automated Dada machine, and might not bother Tzara - or Burroughs, for that matter. Surrealism, however was about both insight and nonsense. It was inspired by invention of photo and film, internal combustion and thoughtful critique of a false sense of civilization.  - JV